Photonics approach to reinforcement learning

Abstract

The role of light is expanding beyond communication and measurement to include computing, as observed in the remarkable progress of artificial intelligence (AI) in the midst of the further digitization of society. This paper reviews the latest developments in photonic research aimed at enhancing reinforcement learning and decision-making tasks, both of which are fundamental functions of AI. We introduce methods for solving the multi-armed bandit problem using chaotic itinerancy of semiconductor laser dynamics and provide an overview of collective decision making that leverages quantum light interference.